Sie sind auf Seite 1von 53





The objective of this chapter is to present a complete

introduction to the macro element method used in modern
crashworthiness. The characteristic feature that distinguishes the
present approach from all other classical formulations in nonlinear
mechanics is that trial deformation functions are postulated on the
basis of experimental observations and not in the form of power or
Fourier series. It is shown how experimentally determined shape
functions are be incorporated into a consistent, mathematically
tractable engineering method. Apart from basic formulation it is
also shown how the macro element concept is implemented in
standard CAD/CAE programs.


Crashworthiness emerged as an extensively explored

engineering field in early 60 following introduction of the safety
standards in the US automotive industry. The new regulations
posed a serious challenge to both practicing engineers responsible
for the development of safe vehicles as well as to scientists and
researchers developing predictive tools. Crash response of
structures involves a number of highly nonlinear phenomena such
as localization of plastic flow, interaction of local and global

buckling modes, large deformations and tearing of material. The
industry driven requirement for adequate computational tools
capable of capturing all the above phenomena triggered the
development of a number of dedicated tools and techniques. In the
early days of crashworthiness experimentation was the main design
tools in the automotive industry. At the same time numerical
techniques, especially FE methods were progressing rapidly
towards highly reliable simulation tools. However, it has been
soon recognized that precise numerical analysis is not always the
only thing the engineer needs. In many cases less accurate,
qualitative answers might suffice or at least help in better
understanding of a given crash phenomenon and assist in planning
of simulation and experimental program.

One of the most successful predictive techniques that

emerged during early days of crashworthiness is the so-called
kinematic approach to large deformation of plastic shells that
originates from the famous work of Alexander published in
1960 [1]. During following decades the kinematic approach,
frequently referred to as a macro element method, has grown to a
consistent, mathematically tractable method of analyzing large
shape distortions of sheet metal structures.

In the macro element method kinematic variables such as

displacement, velocity and acceleration fields are postulated as a
set of trial solutions in variational or extremal formulations of solid
and structural mechanics. The characteristic feature that
distinguishes the present approach from all other classical
formulations in nonlinear mechanics is that trial deformation
functions are postulated on the basis of experimental observations
and not in the form of power or Fourier series. In most cases a
successful formulation of an individual macro element imposes
strict limitations onto the type of a structure that can be modeled by
means of such an element as well as certain restrictions onto the
admissible boundary and loading conditions. Therefore, the range
of applicability of a single macro element is much more narrow
then general formulations of engineering theories like e.g. plate
theory or FE method. However, the advantage of such a restricted
formulation of an element is its simplicity. Frequently solution to
the crushing problem is obtained in a closed form while computer
programs based on the macro element method are fast and do not
require complex input data.

This chapter focuses on the basic formulation and example
applications of the macro element method. Short literature review
pertinent to subjects not covered in this chapter is given in the last

1 The Macro Element modeling concept-illustrative example

The concept of Macro Elements in crash calculations is

explained here on an example of a frontal ship collision shown
schematically in Figure 1.

Figure 1 Ship impact analysis procedure based on dividing the

structure into T X and Y Macro Elements containing plate
intersection lines.

The basic idea of the macro element method is to divide the

structure into a number of representative structural elements and
then determine crushing response of each element separately and/or
an assemblage of interacting elements. In the case of a typical bow
structure of a ship the representative elements are referred to as
X, T and Y sections. Each of the above elements results from
the intersection of two or more major structural plate elements
along a common intersection line. The motivation for dividing the
bow structure into X Y and T elements instead of plate
elements is that the material in the immediate vicinity of a plate
intersection line absorbs most of the impact energy. In an actual
accident the discussed sections are loaded in a variety of ways.
The present illustrative example focuses on an axial crushing
response of X elements when the crushing force is parallel to the
plate intersection line. This type of loading is typical for a side
collision and grounding of ships on a wide rock when response of a
ship structure is governed mostly by crushing of X and T
sections. In addition during an axial loading there is generally a
weak coupling between deformation of distant X and/or T
elements so that in an approximate analysis the deformation of each
element can be treated separately.

1.1 Folding modes of cruciforms (X elements)

Representative folding modes of X elements subjected to

axial loading are shown in Figure 2.

Figure 2 Different folding modes of mild steel 'X' sections tested

under quasi-static and dynamic loading. (Courtesy of the
Impact Research Centre at the University of Liverpool).

Depending on initial imperfections the crushed section may

collapse in a variety of folding modes characterized by distinct
levels of energy absorption. However, a careful examination of
contributing folding mechanisms reveals that all the experimentally
observed deformation patterns can be assembled out of a single,
characteristic deformation mode of a corner element, referred to as
a Superfolding Element (SE). For example the natural folding
mode most frequently observed in experiments, compare Figure 2,
consists of four Superfolding Elements deforming in a symmetric
mode, illustrated in Figure 3a.

Identification of experimentally observed folding modes is

frequently simplified by building models of collapsed elements out
of construction paper, Figure 3b. The paper sheets are

inextensibleso that any deformed pattern that can be made out of a
flat paper by means of local bend lines and without cutting the
paper involves only bending deformations of the shell. Such a
deformation is refereed in the literature to as quasi-isometric
deformation, [6]. On the other hand all the cutouts and openings
that must be made in the paper model to reproduce an actual
folding pattern correspond to localized membrane deformations. It
transpires form the Figure 3b that natural folding mode of an X
element involves both bending as well as membrane deformations.

S t


Figure 3 (a) Symmetric folding mode of a Superfolding Element of the

length L=2c and thickness t. The height of the element 2H
corresponds to the length of plastic folding wave. (b) A
structural model, made of construction paper, illustrates an
exact location of contributing Superfolding Elements within a
deformed X section, [4].

1.2 Internal energy dissipation

Internal energy dissipation in computational models made

of construction paper can be conveniently calculated by using the
concept of localized plastic hinge lines introduced in the classical
Theory of Plasticity. In the hinge line method it is assumed that all
the plastic deformations, either bending, tensile or shear are
localized within the discontinuity lines (regions). For the rigid
perfectly plastic idealization of the material the rate of energy
dissipation is then given as

Eq. 1 E& b = M o& and E& m = N o u&

for bending and tensile deformations, respectively. In Eq. 1

M o = o t 2 / 4 denotes fully plastic bending moment, N o = o t is
the fully plastic membrane force (both quantities are given for unit

length of a plastic hinge), t is the material thickness and o is the
level of plastic flow stress. The doted parameters, & and u& , denote
rates of rotation and tension/compression, respectively. Since the
bending and membrane plastic deformations are now assigned to
separate regions there is no interaction of generalized forces. In
addition it follows from the kinematics of the computational model
that loadings in horizontal bending hinge lines and localized tensile
zones are proportional, refer to Figure 3a. Consequently, the rate
equations, Eq. 1, can be effectively integrated over the complete
crushing process.
The bending angle in upper and lower hinge lines in Figure 3a is
/ 2 while the corresponding angle in the middle hinge is . The
total length of each hinge is 2C, so that, the total dissipation due to
bending is

Eq. 2 Eb = 8CM o

Horizontal fibers in the conical zones of plastic deformations are

elongated proportionally to the distance from the cone apex
(coordinate s in Figure 3a). In the final, completely squeezed (flat)
configuration the apex angle of the conical surface is / 2 , so that,
total membrane energy dissipated in four conical deformation
zones is
Eq. 3 Em = 4 yN o dy = N o H 2 = 4M o
2 t

where H denotes half length of the local plastic folding wave. It

should be noted that parameter H is unknown and must be
determined as a part of the solution procedure.

1.3 Energy balance equation

Energy dissipated in the generalized hinge lines equals the

energy supplied to the system by the axial crushing force acting on
axial shortening of the section, . A convenient measure of the
energy absorption capacity of an element, referred to as a mean
crushing force, Pm , is defined as

eff eff
Eq. 4 Eext = P( ) d = Pm d = Pm eff Pm =
0 0

In Eq. 4 the mean crushing force Pm is given as a mean value of the

external energy functional. It can also be interpreted as average,
specific energy absorption per unit crush of a structure. In Eq. 4
eff denotes the so-called effective crushing distance. Separate
studies show that the aspect ratio of the effective crushing distance
to the length of the plastic folding wave, 2H, equals
eff / 2 H = 0.73 , approximately, and is constant for a wide range of
crushed structures, [6]. The effective crushing distance reflects the
fact that an actual structure cannot be folded completely like a
paper model. There always remains a stack of completely
squeezed plastic lobes that have at least one-order-of-magnitude
higher compression stiffness then an active lobe. Consequently an
actual crushing distance corresponding to the creation of one
plastic lobe is smaller then the length of the plastic folding wave,
2 H . This phenomenon is illustrated in Figure 4 for the case of
completely squeezed square column that developed over a dozen of
plastic lobes.

Figure 4 Section through a completely squeezed square column

illustrates a residual stack of completed plastic folds.

Equating internal and external energies in Eq. 3 and Eq. 4

renders expression for the mean crushing force as a function of
unknown length of the plastic folding wave, 2H.

2H H 4 c M o
Eq. 5 Pm ( H ) = 2 M o +
eff t H t

The function Pm (H ) has a minimum ( Pm / H = 0 ) that provides

for a conservative estimate of the energy absorption capacity of a
cruciform element.

Pm 4 2 C
Mo 0.73 t
Eq. 6
= 2
t t

1.4 Conclusions

The introductory example of the macro element approach

reveals most of the distinctive features of the crushing response of
thin-walled structures. The plastic crushing process develops
following the elastic or elastic/plastic buckling and is characterized
by localization of plastic deformations in relatively small parts of a
structure. Plastic deformations are localized in narrow hinge lines
where most of the plastic deformation takes place while the global
deformation of the structure results from rigid body motion of
undeformed or slightly deformed segments. A typical folding
pattern observed in a variety of structural elements can be
approximated by few folding modes of a corner element.
Deformation of a corner element is described by using the concept
of a specialized macro element referred to as Superfolding Element

The simplified calculation procedure presented in this

section leaves several loose ends in the formulation of the
method. A formal formulation of the macro element method is
presented in the next section.

2 General formulation of the Macro Element approach

This section shows how the formulation of the macro

element approach is built on first principles of structural mechanics
by introducing internal kinematic constrains and by imposing
restrictions onto the set of admissible boundary and loading

2.1 Prerequisites

In the Continuum Mechanics deformation of a body in a

time interval [to, tf ] is defined as a one-parameter family of
configurations (!,)K(V);[to, tf ], in such a way that for each
material point XV the function

Eq. 7 x = ( X, ) [t o ,t f ]

is continuous and has a continuous first and second order time

derivatives, see e.g. [12]. Each invertible and continuously
differentiable1 mapping x = ( X, ) is called a position of the
material point X, XV, at time instant t, t[to, tf], in the
deformation x = (, ) [t o ,t f ] . A region of Euclidean space
corresponding to a current configuration of the body, K(V), is
denoted as = (V , t )

Any mapping x = ( X) X V can be taken as a

configuration (deformed body). In structural mechanics the class
of admissible configurations is typically limited to same special
sub-classes by imposing the internal kinematic constrains. A
specific example of such a constrained class of deformation is the
Love-Kirchhoff shell theory in which it is assumed that the material
fiber that is normal to the midsurface in one configuration remains

The term continuously differentiable which will be used
frequently in this chapter should be understood, unless otherwise
stated, as: continuously differentiable as many times as required
except at some surfaces, lines or points.
normal in any other configuration. Another example is the problem
of plane deformations. In the formulation of any problem in the
field of structural mechanics it is assumed, as a rule, that the region
V and the set K(V) are known a priori.

Once the set of admissible configurations is defined a generic form

of the nonlinear problem of structural mechanics is formulated by
specifying the governing equations. These are:
equations of equilibrium
constitutive relations
compatibility conditions
kinematic boundary constraints, boundary loading and
initial conditions

These equations constitute a set of governing relations for the

unknown deformation ( X, ) [t o ,t f ] . The function ( X, ) ,
which constitutes the solution to the generic problem, will be
referred to as a fundamental solution and denoted as o ( X, ) 2.
Likewise, all other fields corresponding to the fundamental solution
will be identified by superscript 'o'.

2.2 General formulation of the macro element method

The general methods of structural mechanics (e.g. beam or

shell theories) are formulated in such a way that the solution to the
generic problem can be obtained for a wide class of reference
configurations V (undeformed body), boundary and initial
conditions and specific constitutive relations.

In contrast to these methods the macro element formulation

is dedicated to narrow classes of structural elements or even to a
single type of a structural element (or its representative part).
Consequently, sharp restrictions are imposed onto the admissible
reference configuration V that, in the remaining of this chapter, will
be identified with a reference configuration of a macro element.
Typical examples of structural elements modeled by dedicated

In the following considerations it is assumed that such a solution
exists and is unique.
Macro Elements are thin-walled prismatic members (such as the
cruciform sections discussed in the preceding section).
Furthermore, the class of boundary, loading and initial conditions,
for which a given method is designed, is also restricted and in
several cases applies to only one type of boundary conditions. For
example, a macro element that models axial crushing response of
an X section will typically require that one end of the section is
clamped while the opposite end moves with a constant velocity and
remains parallel to the clamped end throughout the entire
deformation. In terms of boundary conditions it means that the
spatial position x u = ( X, ); X S u , [t o ,t f ] of the surface Su
S on which displacements are prescribed is known for each time
instant. Also, the velocity of each point on the surface Su is known
a priori (kinematic loading). When the boundary conditions are
changed usually another version of macro element must be

Under such strict limitations the deformation of a macro

element, Eq. 7, resulting from a strictly defined loading and
boundary conditions can be determined on the basis of
experimental observations as discussed in the previous section. As
a rule, the deformation ( X, ) , is postulated with an accuracy to a
vector of free scalar parameters = [ 1 ,... N ] and constitutes a
set of trial deformations.

Eq. 8 { x * K (V ) : x * = * (V , , ); [t o , t f ]; R 3 }

In order to avoid needless complexity it is further assumed that the

vector does not depend on time. All variables pertinent to trial
deformations, Eq. 8, will be denoted by a superscript '*'.

Once a class of admissible deformations is determined the

general form of the macro element method is formulated in the
following way. First, we note that due to limitations imposed onto
the admissible set of deformations, Eq. 8, the boundary, loading
and initial conditions are automatically satisfied for each
t [t o , t f ] . Second, the kinematic variables i.e. the material and
spatial velocity fields

v * = G ( X, ) g[x( X, ), ] where G ( X, ) ( X, ); [t o ,t f ]

gradient of deformation, F * = x * , and its rate, F& * , are determined
from the admissible deformation fields, Eq. 8, by time and spatial
differentiation, respectively. Therefore, the compatibility
conditions are also automatically satisfied.
All that remains to be done is to solve the equations of equilibrium.
In the remaining of this section we shall consider only quasi-static
deformations. In this case the equilibrium of a macro element, v, at
a time instant t, t [t o , t f ] , can be conveniently expressed via the
following form of the principle of virtual velocities (weak

Eq. 9 E& ext

= E& int

where E& ext

= Tv d ( v) denotes the power input while

E& int
= d dv is the stress power (rate of total internal dissipation
in the case of plastic solids) corresponding to a fundamental
solution. In the preceding expressions T denotes surface traction,
is the Cauchy stress tensor while d is the rate of deformation tensor
(symmetric part of the deformation gradient). In following
calculations the spatial description is used consistently, so that, the
integration in expressions for the total rate of internal and external
work is performed over the current configuration x = (V , t ) . In
the next step of the solution procedure a set of trial functions for
the stress power is established.

&* 1 * *
Eint ( , ) : E& int = v* d dv ; d ij = ( v i, j + v j,i )
* * * *
Eq. 10

Each trial solution in Eq. 10 corresponds to a trial deformation

x * = * ( X, , ) . Since deformations are finite the configurations
(V , ) and * (V , , ) may occupy different regions in
space and therefore the limits of integration in Eq. 9 and Eq. 10 are,
in general, different. The 'discrepancy' between the exact and
approximate rates of internal dissipation is then expressed as a

Eq. 11 R ( , ) = E& int

( , ) E& int
( ); [t o ,t f ]

which is a function of time and a vector of free parameters only.

Since the set of admissible deformations is known a global
residual, R(), which describes 'global accuracy' of each trial
deformation, Eq. 10 can be determined as
tf tf
Eq. 12 R ( ) = E& int
( , )d E& int
( )d
to to

The global residual R() is a function of the vector only.

Consequently, the solution to the problem is reduced to the
determination of an optimal vector of free parameters, ,, which
renders the global residual R() an absolute minimum and thus,
corresponds to the best approximation in the entire space-time
subdomain. Once the vector is known the unknown external
loading can be determined directly from Eq. 9. A more detailed
discussion of this problem is presented in the next section.

2.3 Solution Procedure

This section shows how a solution to the governing

equation of the macro element method, Eq. 12, is constructed.
First, the general procedure for a class of steady state and quasi-
steady state processes is discussed. It is shown that there exists a
class of deformation processes where the upper bound theorem of
the Theory of Plasticity can be applied rendering the condition of
convergence of the kinematicaly admissible solution to the
fundamental solution. Secondly the simple minimum criterion for
the mean crushing force is derived for a class of progressive axial
crushing of prismatic members.

2.3.1 Solution procedure for a steady-state processes

In the energy-time space, {E, }, function

Eint ( ) = E& int
( )d ; [t o ,t f ]

defines a single curve referred to as a fundamental energy

trajectory. Likewise the set of trial functionals

{E *
int ( , ) : Eint

( , ) = E& int
( , )d ; [t o ,t f ], R 3 }
define a set of curves referred to as trial trajectories. All
trajectories start from a common origin, corresponding to the
reference configuration o = (V , t o ) , Figure 5. Typically, the
reference configuration is identified with the initiation of the
crushing process.
Eint ( , ")
E int ( , 2 ) E o ( ) o
Eint ( )

( ,2 )
E int ( , 1 )

( , 1 )
( )
= to = to

Figure 5 a). Fundamental energy trajectory,

Eint ( ) , and trial
Eint ( , ) , with a common origin corresponding
to the reference configuration at t = to. b). A family of non-
intersecting trial trajectories.

In the following we shall restrict the class of admissible

constitutive laws to rigid/perfectly plastic isotropic materials with
convex yield surface and the associated flow rule. In this case all
trajectories are non-decreasing functions of time , or a time-like
parameter, . In the vicinity of a reference configuration, = t o ,
the fundamental and trial trajectories can be expanded into a Taylor

Eq. 13
Eint (t o + t ) = Eint
(t o ) + E& int
(t o )t + O(t 2 )

where the rate, E& int

, defines the slope of a given trajectory at the
reference configuration. The upper bound theorem of the classical
Theory of Plasticity for infinitesimal deformations states that out of
all kinematically admissible velocity fields, defined over the
constant reference configuration, o = (V , t o ) an actual velocity
field minimizes the rate of energy dissipation.

In terms of energy trajectories defined in the energy space it

means that the trial trajectory with the smallest slope at the
reference configuration coincides with fundamental trajectory at
least in the immediate vicinity of that configuration and with an

accuracy to at least first order terms3. In other words the upper
bound theorem defines certain property of the state of the system at
the reference configuration and does not provide any definite clues
as to the impending deformation process. In particular the
principle of minimum work cannot be derived from the upper
bound theorem, at least in the general case. This conclusion is
illustrated schematically in Figure 5, which shows that a trial
trajectory that coincides with the fundamental trajectory at the
reference configuration can intersect other trajectories later in the
deformation process. Thus, the total expenditure of work predicted
by the corresponding solution might be larger or smaller then an
actual work input. Furthermore, a trajectory that is optimal within
the immediate vicinity of a reference configuration may diverge
from the fundamental solution and vice versa a trajectory, which
overestimates an incremental response, may constitute a better
approximation later in the process or may be an optimal
approximation in an average sense.

There are, however, certain sets of trial trajectories for which the
upper bound theorem can still be used as an effective tool in
selecting optimal solution. These are the non-intersecting sets of
trial trajectories illustrated schematically in Figure 5 b. In this case
a trajectory selected on the basis of a minimum slope condition at
the reference configuration remains the best solution later in the
deformation process. The necessary condition for a local minimum
of the slope is

E& int
( , )
= 0

This condition constitutes a set of N algebraic equations for N

unknown components of the optimal vector . A global character
of the minimum is usually demonstrated by referring to a particular
form of the function E& int
. In an abbreviated notation a set of
necessary and sufficient conditions for the global minimum is
denoted as

Eq. 14 o = min[E& int


It can be shown that for a stable deformation process the
curvature of the trajectory at the reference configuration is also
An example of non-intersecting trajectories is a set of straight lines
with a common origin, which describes trial solutions for steady-
state deformations. Typical example of a steady-state deformation
is propagation of a buckle in pipeline or flow of the material over a
toroidal surface during plastic inversion of a cylinder. In steady
deformation, all the process parameters are constant in time at each
spatial location. Therefore, the rate of internal dissipation
calculated, as a volume integral over the instantaneous
configuration of the body, is also constant in time and the
corresponding energy trajectory is a straight line.

All kinematically admissible steady-state processes that

originate from the same reference configuration also result in a set
of straight lines and quite obviously the solution selected on the
basis of the upper bound theorem, Eq. 14 provides for the best
solution in a global sense. So that, the optimal solution to a steady-
state process, o , can be defined from either of the conditions

[ ]
* *
o = min or o
= min lim = min E& int
Eq. 15

Another example of a deformation that can be solved by

applying the upper bound theorem is the quasi-steady deformation
process. This process is defined as a 'perturbed' steady-state
deformation where trial trajectories oscillate periodically around a
straight line that represents a stationary process, Figure 6.

E int ( ", )

Eint ( )

= To
= to tg ( ) =

Figure 6 The quasi steady-state deformation process and a

corresponding family of trial trajectories. An average slope,
, is defined for the full oscillation cycle, To.

In Figure 6 the period of oscillations is denoted as To. The optimal
solution to the quasi-steady deformation is found in a
straightforward way by minimizing the slope at the reference
configuration, Figure 6. In practical calculations, however, the
optimal solution o is usually determined from an average slope
for the full cycle of oscillations


Eq. 16 o = min


In Eq. 16 E is an increment of internal dissipation corresponding

to the full cycle. Obviously, a number of other approximations
based, for example, on the weighted residuals method can be
applied here. Quite surprisingly, however, this possibility seems
not to have received due mention in the relevant literature.

2.3.2 Progressive crushing and global minimum condition

Progressive crushing of prismatic members is a special case of the

quasi-steady deformation. It has received a significant attention in
the literature due to its applicability to energy absorbing devices.
The crushing of prismatic members is characterized by the
presence of highly localized zones of plastic flow. Outside these
zones the shell is assumed to be rigid. The rigid parts, however,
can be subjected to an arbitrary rigid motion characterized by a
rigid-body translation vector, and rigid-body rotation vector . In
the rigid-body dynamics, the external loads are the global cross-
sectional forces, P, and moments, M. Consequently, the rate of
external work is

Eq. 17 E& ext

= P& + M&

and in general consists of six contributions.

In the next section progressive axial crushing of prismatic

members will be discussed as a representative example of the
application of the macro element method. In the case of axial
crushing it is convenient to use an actual axial shortening of the
member, as a time-like parameter of the process. Furthermore,
for a perfectly/plastic material we can assume, without a loss of
generality, that the rate of deformation & is constant throughout
the deformation process. Consequently, multiplying the time

coordinate by a constant velocity, & , does the transition from the
energy/time space {E, } to the energy/time-like parameter space
{E, }. This transformation changes the shape of trajectories in
Figure 6 but does not affect relations between their average slopes.
So that, the minimum condition, Eq. 16, can be rewritten in the


Eq. 18 = min

where is an axial shortening corresponding to a full cycle of

oscillations or to the creation of a single plastic lobe. Hereinafter,
will be identified as an effective crushing distance in axially
crushed prismatic members. During an axial crushing the total
power input to the system is due to the axial crushing force, P,
acting on a conjugate rate of axial shortening & , E& ext *
= P& . The
total expenditure of work corresponding to a full cycle of
oscillations = ( t o + To ) ( t o ) , is simply Eext
= Pm and
* *
equals to the total internal dissipation Eint . Substituting Eext into
Eq. 18 finally predicts


Eq. 19 o = min = min[Pm ]

The above result is simply the criterion of an absolute minimum of

the mean crushing force, Pm, per single plastic lobe. This criterion
was conjectured rather then proved by Alexander in 1960, [1], and
has been used ever since in virtually all solutions concerned with
progressive crushing although up to now the conjecture has never
been substantiated by a rigorous proof.

2.3.3 Possible generalizations of the global minimum


In the preceding sections it has been shown that the solution

to the progressive crushing problem can be obtained from the upper
bound theorem in the case of a particular set of non-intersecting
energy trajectories, Figure 5. The question remains, however,
whether an actual deformation is governed by the requirement of an
absolute minimum of the total expenditure of work. Deeper insight

into this problem is obtained from the analysis of the straining
history of a representative material point.

In 1986 Hill, [13], proved that there exist optimal paths of

homogenous deformation between different states of finite strains
for which the total expenditure of work gives a global minimum.
An optimal path is such that certain triad of material fibers is
perpetually orthogonal, while the logarithms of their stretches vary
monotonically in fixed ratio. In other words optimal paths
correspond to a pure straining deformation which, moreover
coincides with a path of proportional loading in the space of
logarithmic strains

log1 log2 log3

Eq. 20 = = =t
&1 &2 &3

In Eq. 20 i denotes i-th component of the principal stretch while

&i is the rate of logarithmic strain, &i = &i / i . In a plane-strain
deformation of incompressible material the principal Eulerian
strain-rates are always automatically in fixed ratio, namely {1, -
1, 0}. Hence, the condition of proportional loading, Eq. 20, is
identically satisfied. Therefore, any (locally) plane pure straining
deformation coincides with an absolute minimum of the total work
expenditure (for given initial and final strains).

There is strong experimental evidence that in fact the above

conditions are fulfilled in a majority of representative deformation
mechanisms associated with local collapse of crushed shells.
Indeed, the deformed pattern of thin-walled members can be
assembled out of axisymmetric shells such as moving and/or
stationary cylinders, cylindrical cones and toroids. Moreover, it is
observed that plastic deformations are negligible in one of the
principal directions and therefore majority of computational models
assume inextensibility in this direction. Such an axisymmetric
deformations are 'locally plane' in planes that are perpendicular to
the direction of inextensibility. Since during such deformations the
same material fibers are principal throughout the crushing process
both optimum conditions are satisfied. This implies that an actual
deformation of crushed shells follows the path of minimal work

Even more importantly the above conclusion remains valid for a
strain-hardening materials as well as for a purely geometrical state
Eq. 21
&12 + &22 + &32 dt

The path integral in Eq. 21 is frequently identified with an

equivalent strain (or is proportional to such a strain measure). This
corollary to the minimum condition has an important implication: if
the requirement of an absolute minimum of work is an underlying
law that governs crushing response of shells then elements made of
highly strain-hardening as well as non-metallic materials should
deform in a manner similar to elements made of plastic material.
This supposition has strong experimental evidence. It is observed
that typical deformation patterns of shell-like structures made of
sufficiently ductile materials can be approximated, with a
reasonable accuracy, by segments of axisymmetric shells. This
simple geometrical observation suggests that regardless of a
material property crushing deformation of shells tends to follow an
optimal path defined by Eq. 20. Therefore the criterion of the
minimum mean crushing force is also used in the case of shells
made of strain hardening as well as non-metallic materials. The
only limitation here seems to be a sufficient ductility of material, so
that, large strains in localized deformation zones can be
accommodated without damage of material or rupture of a shell.

3 The Superfolding Element (SE) method

In the preceding section it has been shown that the crushing

deformation of shell-like structures results from a local loss of
stability and creation of the so-called local plastic fold or plastic
wave. Once created, the plastic fold accommodates most of the
plastic deformation in a shell. The local deformation process
continues up to the point where local contacts prevent further
deformation of an actual fold and induce development of
subsequent fold. Such a deformation process is referred to as a
progressive crushing or progressive folding process.

An interesting feature of the progressive folding is its

'geometrical similitude'. It has been observed that most of actual
deformation patterns can be assembled from a single, typical

folding lobe. A single Superfolding Element (SE) models the
crushing behavior of such a lobe. This section provides for an
overview of underlying concepts of the Superfolding Element
method. The presentation starts from general formulation of the SE
used in computerized applications. Then, detailed discussions of
possible simplifications to the general solution are presented. All
results of this section are valid for shells made of plastic isotropic
strain hardening materials with the convex yield surface and an
associated flow rule.

Figure 7 A checkerboard of SE in a spot welded hexagonal column

illustrates the discretization procedure by using SE's. The
insert reveals the 'macro-size' of a SE as compared to the
standard FE mesh, [8].

3.1 Discretization into Superfolding Elements

In the initial undeformed configuration a SE represents the segment

of a corner line of a prismatic column, refer to Figure 7. It is cut
off from a column by a set of two parallel horizontal planes. The
distance between planes, 2H, equals the length of the plastic
folding wave of the column. The vertical boundaries of a SE are
defined by a set of two vertical planes equally distanced from the
neighboring corners and/or vertical edges of a column.

3.1.1 Dimensions of a SE

In the initial, undeformed configuration a single SE is

defined by four parameters:
1. total length, C, of two arms of a SE, C = a + b,
2. central angle,
3. wall thickness ta of the arm of the length a
4. wall thickness tb of the arm of the length b

Figure 8 Basic cross-sectional dimension of a Superfolding Element

shown on the insert in Figure 7.

In stamped sheet metal structures ta = tb = t while in extruded

aluminium components thickness of the two adjacent walls might
be different.

It should be noted, at this point, that the height of a SE

which corresponds to the length of a plastic folding wave, 2H, is
not given a priori and must be calculated as a part of the solution.
Accordingly, a computerized implementation of the SE method
requires an adaptive meshing algorithm.

3.1.2 An active layer of folds

A set of Superfolding Elements located between two horizontal

planes defines a single layer of plastic folds (also referred to as a
deformable cell, Figure 9).

Figure 9 A deformable cell represents a single layer of active plastic
folds in a progressively crushed prismatic column, [8].

The number of SE in a given layer corresponds to the number of

corners in a column. In progressive crushing of real columns
plastic deformation are always spread over two neighboring layers.
However, consideration of a single layer at a time is a useful
approximation, which leads to accurate results. An example of
possible deformation patterns of a single active layer for various
regimes of loading is shown schematically in Figure 10, [8].

Figure 10 Various deformation modes of an active layer of folds

in a square prismatic beam subjected to different loading
histories. The symmetric mode (b) corresponds to the
progressive axial crushing process, [8].

The present section is concerned with axial loading only. In

this case the deformation of all SE is symmetric and the boundary
planes remain parallel throughout the deformation process.

3.2 Folding modes of a SE

The most general deformation mode of a single SE is shown

schematically in Figure 10. The plastic folding of the element
involves activation of five different deformation mechanisms.
These are:
continuous deformation of a section of the floating toroidal
surface, 1, at the so called corner point
bending along horizontal stationary hinge lines, 2,
rolling deformations along moving inclined hinge lines, 3,
extensional deformations of the conical surface, 4, in the
terminal phase of the deformation process and
bending deformations along inclined hinge lines, 3, in the
terminal phase of the folding process, when the moving
hinge line is locked within the element.

Figure 11 Basic folding mechanisms in a deformed Superfolding

Element (SE):
Deformation of a floating toroidal surface.
Bending along stationary hinge lines.
Rolling deformations.
Opening of a conical surfaces.
..Bending deformations along inclined, stationary, hinge
lines following locking of the traveling hinge line .

The general, folding mechanism in Figure 11 is constructed

from two simpler folding modes, illustrated in Figure 12. These
modes are referred to as an asymmetric and symmetric deformation
mode, respectively. The mode shown in Figure 11 is called an

asymmetric mixed mode. A progress of the deformation process in
each mode is controlled by a single process parameter ,
0 f , which defines the rotation of a side face of an element
from the initial upright position, Figure 11. At the initiation of the
folding process = 0. The process terminates when = f =/2.
The asymmetric deformation mode is characterized by the lack of a
conical surface 4 in Figure 11. Consequently, the propagating
hinge line, 3, controls the entire folding process. The symmetric
deformation mode, on the other hand, lacks the propagating hinge
line 3 in Figure 11. In this case local extensional plastic
deformations are confined to the conical surface 4 as discussed in
section 1.

The development of a particular folding mode in Figure 11

and Figure 12, is controlled by a single switching parameter
* , 0 * f . This parameter defines a configuration at which
a symmetric mode takes over the control of the folding process. If
* = f the folding of a SE is controlled by an asymmetric mode
alone while the case * = 0 corresponds to a purely symmetric
mode, see Figure 12. For 0 * f both mechanisms are
involved in the folding process: folding starts as an asymmetric
mode and continues up to the point where the moving hinge line 3
is locked within an element. At this point the conical surface 4
starts to grow.

Figure 12 Two fundamental folding modes of a single

Superfolding Element controlled by limit values of the
switching parameter *.

An actual value of the switching parameter, *, depends on
both the input parameters, {C, t, }, and constraints imposed onto
deforming faces of a SE. In the case of an unconstrained or
standing alone SE the asymmetric mode of deformation, * = 0, is
predominant for right angle and acute elements, / 2 , while
the symmetric mode controls the folding process of obtuse
elements with the central angle, larger then 120 degrees,
approximately. In the intermediate range of central angles both
modes coexist while the fractional contribution of each mode to the
total energy dissipation depends on the central angle, , and the
width to thickness aspect ratio, C/t. The folding modes of a
standing alone SE are referred to as natural folding modes.

A SE, which is a member of a deformable cell, is

constrained by neighboring elements. Kinematic constraints are
introduced into the element either by imposing the deformation of
one or two arms of the element in a pre-defined direction or by
constraining the deformation of the element's corner line. The
former case is typical for an assemblage of elements, which model
the deformation of a column with closed cross-section. In this case
the requirement of the circumferential continuity of the
deformation field may force an element to the deformation mode
different then the natural folding mode. Similarly, constraints
imposed onto the corner line can change the natural deformation
mode of an element. During the deformation of 'X' and 'Y'
sections, discussed in earlier in this section, the continuity
conditions imposed onto the common corner line of all contributing
flanges prevent the development of asymmetric modes.

3.3 Trial functions

An example of a kinematicaly admissible configuration,

x = ( X, , t ); t [t o , t f ] , of a single SE is shown schematically
* *

in Figure 11. In general vector has four components.

Eq. 22 o = {r , H , * , eff }

These are (refer to Figure 11):

rolling radius, r,
length of the plastic folding wave, 2H,

switching parameter, *
effective crushing distance, eff.
It has been shown in section 1 that eff / 2H = 0.73, approximately,
for all progressive crushing processes of practical importance.
Consequently, the effective number of free parameters of the
process is reduced to three, = {r, H, *}. The velocity field, v*,
and the corresponding rate of deformation tensor, d*, in Eq. 10 can
now be calculated on from the postulated deformation,
x * = * (V , , ); [t o , t f ] , by an appropriate spatial and time
differentiation. The calculated trial fields are then used to construct
trial solutions Eint discussed in section 2.2. In the next section the
trial dissipation Eint for a SE will be expressed in terms of the rate
of curvature and rate of extensions defined at the middle surface of
the element. Particulars of the transition from a three-dimensional
description of continuum to a two-dimensional formulation of a
shell element (in Eulerian description) are not discussed here.

3.4 Trial solutions

The rate of internal energy dissipation in a deformed shell

element results, in general, from the continuous and discontinuous
velocity fields, [2].

Eq. 23
[ ]
E& int = (M & + N & )dS + i =1 i M oi &i dl i

In Eq. 1 S denotes the current shell mid surface, n is the total

number of plastic hinge lines, Li is the length of the i-ht hinge
while [& ] denotes a jump of the rate of rotation across a moving
hinge line. Components of the rate of curvature and rate of
extensions tensors are denoted, respectively, as & and & while
M and N are the corresponding conjugate generalized
stresses. For the sake of simplicity in the remaining part of this
section the superscript '*' is omitted in all expressions for trial
fields. The arguments and will be specified whenever
Since the assumed deformation fields are axisymmetric and, in
addition, inextensible in one of the principal directions, tangent to

the shell mid surface, the expression for (trial) rate of internal
energy dissipation, Eq. 23, reduces to

Eq. 24
[ ]
E& int = N o &1 dS + i =1 i M oi &i dl i

where &1 is the rate of straining in a principal direction, tangent to

the shell's mid surface and equals to a corresponding component of
the rate of deformation tensor &1 = d11 . The strain rate component
perpendicular to the mid surface does not contribute to the internal
energy dissipation due to specific form of the Tresca yield
condition used in the present calculations. Integrating Eq. 24 in the
interval 0 f renders the expression for (trial) energy
dissipation per single plastic lobe
* f
Eq. 25 Eint ( ) = (1) E& int d + * (2) E& int d

Two integrals on the r.h.s. of Eq. 25, defined through the switching
parameter *, correspond to the contribution of asymmetric and
symmetric mode, respectively.

In practical calculations it is convenient to define following

expressions for the membrane and bending contributions to the
internal energy dissipation, [4], [5], [6]

Eq. 26

EintN d N 0&dS = 0N ( ) d t&dS
0 S 0 S

[ ] n t
i =1 i M oi &i dl i = i =1 ( i ) 0M ( ) d i &i dl i

40 L
[ ]

where oN and oM denote, respectively, an average level of the

flow stress in the entire crushing process. This stress is referred to
as an energy equivalent stress. The functional dependence of the
energy equivalent stresses on the constitutive relation and
characteristic deformations of the SE is discussed in the next

3.5 The energy equivalent stress measure

Determination of proper flow stress measures is crucial for

the accuracy of all calculations based on the macro element
approach. Although the stress state in each zone of localized
plastic flow is typically one-dimensional the determination of the
level of plastic flow is not a straightforward task and require right
understanding of the global approach used in macro element
method. This section introduces the global or average measures
compatible with dissipation energy measures defined in the
preceding section.

3.5.1 Quasi-static processes

The material properties enter the energy relations, Eq. 25,

via the equivalent stresses oM and oN. These stresses are defined,
respectively, as

0 N ( ) =
( )d
Eq. 27
0 M ( ) = ( ) d

2 0

and correspond to an average level of the plastic flow stress in

regions subjected to quasi-static uniaxial tension/compression or
bending, characterized by a representative strain (in most cases
corresponds to maximal strain in a given region). In Eq. 27 ( )
corresponds to a standard quasi-static tensile characteristic
(detailed derivation of the above formulas is given in [4]).

3.5.2 Dynamic processes

The visco-plastic constitutive relation, ( , &) , is

postulated as a product of two contributions

Eq. 28 ( , & ) = v ( , &v ) ( , & )

where v () corresponds to the standard quasi-static tensile
characteristic of a given material, determined at the constant strain
rate, &v , (10-3 [1/s] &v 10-4 [1/s]) while the dynamic
factor () describes strain rate effects. As usual the stress and
strain measures, used in the Eq. 27 and Eq. 28, correspond,
respectively; to the Cauchy stress and logarithmic strain .

3.5.3 Determination of energy equivalent measures from

standard material tests

Correct definitions of dynamic stress and strain rate

measures require additional processing of standard experimental
data. The general processing procedures are explained here on an
example of a simple power-type-hardening law. These methods are
then applied to an actual stress-strain rate data obtained from
laboratory experiments. Quasi-static energy equivalent flow stress.

Consider a simple strain-hardening material defined by the

power-type-hardening law:

( )
Eq. 29 =
y o

In Eq. 29 y corresponds to a 0.2% proof stress (o= 0.002), is a

current strain while n corresponds to stress hardening exponent. In
a plastic deformation process the density of energy dissipation, e,
corresponding to an actual level of monotonically increasing
straining, , is given as:

Eq. 30 e = ( ) d

which together with Eq. 29 predicts:

Eq. 31 e= ( ) or e = ( )
n +1
In Eq. 31 the energy equivalent flow stress ( ) is defined as

Eq. 32 ( ) = y = ( )
n + 1 o n + 1

A comparison of both above defined stress measures is shown in

Figure 13.

Figure 13 A power-type-hardening material (n=0.2), solid line

and corresponding energy equivalent flow stress, broken
line. Dynamic energy equivalent flow stress

In a logarithmic scale the graphs of energy density, Eq. 31

and energy equivalent flow stress, Eq. 32, correspond to straight
lines. Assuming that this feature of the energy density and
equivalent flow stress function holds also in the dynamic case using
a CowperSymonds-type of constitutive relation, conveniently
captures the strain-rate effect:

1/ p
e d ( , &) &
Eq. 33 e d ( , &) = e( ) f (&) or = 1+
e( ) D

where D and p are two new material constants responsible

exclusively for the rate effects. In Eq. 33 & is the rate of strain
while the relation e( ) is a fictitious static energy density function
corresponding to & = 0 . In practical applications e( ) is identified
as an energy density corresponding to the quasi-static tensile test.
Once the energy density function is defined by means of the second
of Eq. 33 the corresponding dynamic energy equivalent flow stress
is calculated as:

e( ) & & 1 / p
1/ p

Eq. 34 ( , &) =
1 + = ( ) 1 +

So that, in the case of linear approximation to the strain rate effect

the dynamic factor () in Eq. 29 has a form
1/ p
Eq. 35 (&) = 1 +

And does not depend on an actual strain level Illustrative example

In the case of an actual strain-stress relation deviations of

various stress measures form a straight line can be quite significant.
However, experimental observations show that typically the
dissipation density function, Eq. 31 is close to a straight line (in a
logarithmic scale), as illustrated in Figure 14 for typical mild steel.

Figure 14 The experimental energy density function for a mild

steel (solid) and corresponding linear approximation (quasi-
static response).

So that the definition energy equivalent flow stress relations, Eq.
27, and material constants D and p in Eq. 34 is done in three steps:
First, the energy density functions are calculated for all
experimental dynamic and quasi-static tensile curves.
Second, the linear approximations to these curves are defined
by means of linear regression method. An example of the
corresponding procedure is shown in Figure 14.
Finally the D and p values are calculated by applying the linear
regression method to the slopes of energy density functions at
various deformation rates. This procedure is illustrated in
Figure 15 for the reference strain equal to 20%.

Figure 15 Dynamic energy density function normalized with

respect to the quasi-static response as a function of the
strain rate. Energy density is calculated for a reference
strain of 20%. Beaded line corresponds to experimental
findings while the broken line shows results of the linear
regression method.

It should be noted in passing that the strain rate constants D

and p calculated by means of the above-discussed procedure are
quite different then the values reported in the literature for the
initial yield of material. For example the typical values for the
initial yield of a mild steel are p=5 and D=40.4 [1/s] while in the
case of energy equivalent flow stress the corresponding data are
p=7.5 and D=5466000 [1/s].

4 Crushing response of simple structures

The explicit expression for the mean crushing force, Pm,

corresponding to a complete folding of a standing alone SE is
derived by substituting expressions for contributing energy
dissipation mechanisms, Eq. 26, into the governing minimum
condition, Eq. 19. The complete derivation of the following result
is given in [4]. The final form of the governing expression is

t2 r C H
{ o N ( 1 ) A1 + o ( 2 ) A2 + o ( 3 ) A3 +
Pm =
4 t H r
H 2 H
+ o ( 4 ) A4 + o ( 5 ) A5}
Eq. 36
t eff
where Ai = Ai (, * ), i = 1, 2 ...5

The five terms in parenthesis on the right hand side of Eq. 36

describe, respectively, fractional contributions to the total energy
dissipation resulting from five elementary deformation
mechanisms, identified in Figure 11. The five factors, Ai i = 1,2...5
result from the surface-time integration. Factors. A2 and A4, are
easily calculated as a closed-form functions of geometrical
parameters. The remaining factors A1, A3 and A5, are functions of
elliptic integrals and must be calculated numerically. The meaning
of other variables appearing in Eq. 23 is explained in section 3.1.1.

Parameters corresponding to the equilibrium of a SE are

determined from the set of three nonlinear algebraic equations,
resulting from the minimum condition, Eq. 19

Pm Pm Pm
=0 ; =0 ; =0
Eq. 37 H r *

In general the set of governing equations, Eq.'s 24, has no

closed-form solution. Such a solution exists, however, for two
fundamental folding modes of a Superfolding Element illustrated in
Figure 12. Such a solution will be discussed later in this section in
conjunction with the crushing behavior of square columns.

4.1 Single Superfolding Element

Equation 23, can be easily generalized to the case of a

Superfolding Element with various thicknesses, ta and tb, of two
arms, a and b respectively, refer to Figure 8.

Eq. 38
1 2 N r at a2 bt b2
Pm = {t1 o ( 1 ) A1 + A2 ( o ( 2 ) + o ( 2 ) )+

4 t1 H H
H 2 M H 2H
t 32 o ( 3 ) A3 + t 4 o ( 4 ) A4 + t 52 o ( 5 ) A5}

r t4 eff

Such a cross-section is typical for extruded aluminium elements

where the difference in thicknesses of neighboring walls can be as
large as one hundred percent or even more. The plastic energy
dissipation in consecutive folding lobes depends now on the
direction of folding of the SE. For example, in the case illustrated
in Figure 11 the moving hinge line 3 sweeps the face a of the
element and, thus, involves the deformation of a shell of the
thickness ta. At the some time the conical surface 4 is developed at
the boundary between both faces and propagates into a thinner
face. This feature of the folding process is reflected in the
governing equation, Eq. 38, by specifying an appropriate thickness
for each contributing mechanism. For example, the deformation
pattern in Figure 11 corresponds to the following set of thicknesses,
ti i = 1, ...5:
t1 = t3 = t5 = ta and
t4 = ta for ta tb or t4 = tb for tb < ta.

Similarly as in the case of a uniform-thickness element the

equilibrium solution to the Eq. 38 is obtained via the minimization
procedure, Eq. 24. It should be noted, however, that in this case a
useful approximation of the folding mode by pure asymmetric or
symmetric folding modes does not apply, even in the case of a
right-angle element, and consequently, the solution must be found
by minimizing the complete governing equation, Eq. 37.

4.2 Crushing response of an assemblage of Superfolding

The crushing response of a single layer of folds, compare

Figure 9 and Figure 10 is calculated by summing up fractional
contributions of all SE in a given layer. Accordingly, the
governing equation of the problem is

Eq. 39
ti2 N r C H
Pm = { 0 ( 1 ) A1i + o M ( 2 ) A2i i + o M ( 3 ) A3i +
i =1 4 ti H r
H 2H
+ o M ( 4 ) A4i + o N ( 5 ) A5i }
ti eff
where the summation is extended over the J contributing SE. It is
assumed here that the column is made of one material, so that, all
average stresses are calculated on the basis of a single constitutive
relation. Each element, however, may have different geometrical
dimensions: Ci, i and ti, i = 1, 2, .....J.

Since, all elements in a given layer of folds deform with the

same length of the folding wave, 2H, there is only one 'H'
parameter in the governing equation. In order to simplify the
calculation routines it is also assumed that values of the two other
free parameters, i. e. the rolling radius r and the switching
parameter * , are also the same in all contributing elements.
Consequently, the solution procedure for an assemblage of SE
parallels the corresponding procedure for a single SE.

4.3 Closed form solution for rectangular columns

Theoretical procedures developed in preceding sections are

applied here to the crushing response of a rectangular column of
uniform thickness, t, and cross - sectional dimensions 2a and 2b,
Figure 16.

Figure 16 Cross-sectional dimensions of a rectangular column.

The column is made of a strain hardening material defined by the

power - type constitutive law, Eq. 29. The governing equation of
the crushing problem, Eq. 39, can now be rewritten in a simplified
form, applicable to right - angle elements, [2]

t2 1 r C H 2H
Eq. 40 Pm = (o A1 + o2 A2 + o3 A3 )
4 t H r eff

where oi, i = 1,2,3, denotes equivalent stresses in three main

regions of plastic deformations, refer Figure 11, while C is the
length of a representative SE, C = (a+b), refer to Figure 16. For a
power-type constitutive relation, Eq. 29, energy equivalent stresses,
oN and oM are given as
0N ( ) 1
y 1 + n 0

o M ( ) 2
Eq. 41 =
y (n + 1)(n + 2) 0

where denotes the characteristic final strain. In the case of an

uniform tensile-compression deformations corresponds to the
final strain measured at any point of the deformed region whereas
in the case of bending deformations equals to the final strain at
an outer fiber

The relevant strains for three main mechanisms of plastic

deformation in Eq. 40 are readily determined from the kinematics
of the folding process. Consider first rolling deformations. The
final strain at the outer fiber of a traveling hinge line 3 Figure 11 is
inversely proportional to the rolling radius, r, and equals

1 = ln(1 + )
Eq. 42 2r

Similarly, the final strain at the outer surface of horizontal

stationary hinge lines is
2 = ln(1 + )
Eq. 43 2R

where R is the large radius of the toroidal surface. It follows from

the kinematics of the process and in particular from the formula for
the effective crushing distance that radius R equals R= 0.54H
where 2H is the length of the plastic folding wave. Thus, the
representative strain in the region 2 Figure 11, equals
= ln(1 + 0. 93
2 )
Eq. 44 H

Determination of a compressive strain in the moving

segment of a toroidal surface, 1, in Figure 11 involves complex
calculation, which exclude the possibility of getting a desired
closed-form solution. However, separate numerical and
experimental studies show that strains in this region are of the same
order as those in moving hinge lines 3, [6]. Therefore, in the
following analytical calculations Eq. 44 is used as a convenient
approximation of final strains at material points on the toroidal

Further simplifications to the expression for a characteristic

strain, , essential in analytical calculations, are achieved by
expanding Eq. 42 and Eq. 44 into power series

t t t
1 2 0. 93 3
Eq. 45 2r H 2r

where only the linear terms are retained. Substituting Eq. 45 back
into the expressions for energy equivalent flow stresses, Eq. 41,
finally yields
1 0.5
1 o N
( 1 ) = y a1 a1
r n + 1 y

2 o M ( 2 ) = y a 2
2 0.93
(n + 1)(n + 2) y
2 0 .5
Eq. 46 3 o M
( 3 ) = y a3 a3
r (n + 1)(n + 2) y

Having determined energy equivalent stresses, i, as

functions of kinematical parameters, r and H, the governing
equation, Eq. 40, can be rewritten in the form

t y
r 1n C t
1+ n
H t 2 H
1+ n

Pm = A1 + A2 + A3
4 t t H t r eff

Eq. 47 Ai A a
i i

where constant coefficients Ai, i=1, 2, 3, are functions of the

central angle of a SE while coefficients ai, i=1, 2, 3 , defined in Eq.
46, are functions of material parameters only. For example, a set of
coefficients Ai pertinent to the crushing response of a SE with the
central angle #=90o is:
A1 = 8 I1 I1 = 0.555
A2 =
Eq. 48 A3 = 2 I3 I3 = 1.148

The equilibrium of the system, governed by Eq. 47 is

defined by applying the minimum conditions, Eq. 37. The final
results for optimal free parameters of the process, i.e., length of the
folding wave 2H and rolling radius r, are, [4]:
2+ n 1 1+ n
C 3+ n
= A 1
3+ n
A A 3+ n
3+ n
3 K 1 t ;

K 1
= (1 + n)(1 n) 3+ n

1+ n 2

1 n
C 3+ n
= A 1
3+ n
3+ n
2 3
3+ n
K 2 t ;
1+ n

Eq. 49 K 2
= (1 + n)(1 n) 3+ n

Substituting Eq. 49 back into Eq. 47 renders the closed-form
solution for the average crushing force Pm.
1 n
(1+ n ) 2 1 n 2
Pm (1 n )
C 3+ n
= A 1
3+ n
3+ n
A 3
3+ n *
Eq. 50
My t

* [K 1 n
+ K2
(1+ n )
+ K 2 K1
(1+ n )

In Eq. 50 the fully plastic bending moment, My, is

calculated with respect to the yield stress of material, y , and
equals, My = yt2/4. For the perfectly-plastic material, n = 0,
results given in Eq. 49 and Eq. 50 converge to the solution for a
perfectly plastic material.

H 3 A2 3 C 2
= ; C = (a + b)
t A1 A3 t
r A A C
= 3 2 2 33
t A1 t
Pm C 2H
= 3 3 A1 A2 A3 3
Eq. 51 My t eff

which for given values of constants Ai, Eq. 48, and a given
non-dimensional ratio of the length of the plastic folding wave to
the effective crushing distance, 2H/eff 1.37, finally yields.

H 3 C

() 2

C = (a + b)
r C
= 0. 715 3

Eq. 52 t t
Pm C
= 13. 052 3
My t

The later result is valid for a single right - angle element

and differs form the result for the entire column in the level of
non - dimensional mean force, Pm/My. For four SE contributing to
the crushing strength of a rectangular column the mean crushing
force is simply four times larger then the resistance of a single SE
and equals, [5]
Pm C
= 52. 2 3
My t
Eq. 53 C = (a + b)

while the length of the plastic folding wave, 2H, and the magnitude
of rolling radius, r, are the same as in the case of a standing alone

4.4 Dynamic response

In the constitutive relation, Eq. 35, the dynamic energy

equivalent flow stress is given as a product of quasi-static stress
and dynamic factor that depends on the strain rate only. Thus, in
the present formulation the strain-hardening and strain-rate effects
are effectively decoupled. This feature of the constitutive relation
greatly simplifies calculations for dynamic crush. Indeed, for a
dynamic crushing process characterized by a constant strain-rate & ,
the integration of energy equivalent stresses, Eq. 26, reduces to the
multiplication of the quasi-static average stress by the dynamic
factor (&)

0 N ( , &) = ( ) (&)d = (&) N
0 ( )
Eq. 54
0 M ( , &) = ( , &)d = (&) 0M ( )

2 0

Consequently, the dynamic crushing force, Pmd , is also calculated

by multiplying the quasi-static force, Pm, by the dynamic correction
factor, (&) .

The average rate of strain, & , is calculated from the known

kinematics of the folding mechanism and assumed constant
velocity, & , of compression of a single lobe. The average strain
rates depend on the deformation mode of a SE and equal

& 0.33
Eq. 55 & 0.43
& 0.25
for the asymmetric, mixed and symmetric modes, respectively
(details of the derivation procedure are given in [5]). In Eq. 55 the
velocity of axial compression & must be given in meters per
second, [m/s], while C (the total cross-sectional length of a
representative SE) is given in meters, [m]. For a cross-section
composed of more then one SE the representative length, C, should
be calculated by dividing the entire cross-sectional length, L, by the
number of contributing elements, J.

Having determined a representative level of the strain-rate

the mean dynamic crushing force, Pmd , is calculated from the

& 1 / p
Eq. 56 P = Pm 1 +

where Pm is the quasi-static mean crushing force defined Eq. 50.

5 Design and calculation of energy absorbing systems

Design of energy absorbing structures is achieved in several

steps, usually in a highly iterative design/calculation process.
Objectives of each design step are secured by usage of dedicated
computational tools typically different for each step. The selection
of an appropriate tool for a given step depends on the complexity of
the problem and availability of suitable software or other predictive
techniques such as engineering formulas, empirical data or
handbook - type of information. For example, in the pre - design or
pre - prototyping stages factors such as the specific energy
absorption per unit length or the maximal moment capacity of a
cross - section are of primary importance. In the case of typical
cross - sections such information is readily available in handbooks.
In the case of more complex, real - word, members the desired
parameters can be easily calculated by using specialized tools, such

Using of commercial FE packages at early stages of design

process are not justified due to the excessive modeling effort. On
the other extreme, only professional FE packages such as
PAM-CRASH, DYNA or RADIOSS guarantee the detailed, full
crash simulation. In the intermediate stages of the design process
usage of both computational tools is highly desirable and can
significantly increase the efficiency of the design process.
Example of such a process, typical for automotive industry, is
shown schematically in Figure 17. This figure also shows a chart
illustrating possible application of various predictive tools at
different stages of the simulation based design process.

(a). (b).

Figure 17 (a). Five stages of the design process leading to a

crashworthy system. (b). Areas of applicability of various
tools in the simulation based design environment.

This section shows how the concept of Superfolding Element,

implemented in the program CRASH CAD, is used at various stages
of the design process. It is shown how synthetic approach to the
design process proceeds from selection of a cross - section and its
basic dimensions, through component design into the design of
structural assemblages and culminates in full FE crash simulation.

CRASH CAD is a highly interactive CAD/CAE software based on
the macro element approach introduced in the preceding sections.
5.1 Crash analysis at the level of a cross - section.

The design of a structural member at the cross - sectional

level is especially important at the pre - design and early design
stages when the proper shape (topology) and optimal dimensions of
a member are sought and the design concept undergoes frequent
modifications. Application of CRASH CAD at this level is especially
attractive as the program requires as input only overall dimensions
of the cross section, and tensile characteristic of the material
while the calculation process takes only few seconds on a standard
PC. Consequently, designer can examine a wide range of
cross - sectional topologies and run several parametric studies
within only few hours of work. The simplicity of CRASH CAD
input is illustrated in Figure 18 for a typical hat cross - section
made of aluminium alloy.

Figure 18 Input data screen of Crash Cad with a discretized

cross - section of a hat member. Only overall dimensions of
a section are needed in order to define the input data file.

Ranking of various cross - sectional topologies is shown in

Figure 19 for a number of aluminium sections tested at Alcan
laboratories, [7]. All sections were made of a 2-millimeter thick
AA5754-0 aluminium sheets and tested under the axial crash
loading. For the sake of comparison Figure 19 shows both
experimental data and CRASH CAD predictions for the average
crush force (specific energy absorption capacity per unit length)
and length of the plastic folding wave two; factors of primary
importance for the optimal design of sections against crash.

2 mm AA5754-O

Average Crush CRASH CAD 28.1 53.2 41.9 66.6

Force (kN) Experiment 26.0 49.6 39.8 70.1

Fold Half CRASH CAD 35.9 32.4 36.0 46.1

Wavelength (mm) Experiment 37.5 33.5 36.5 41.5

Figure 19 Ranking of various cross-sectional topologies from the

point of view of energy absorption capacity. The comparison
between experimental and theoretical data emphasizes a
typical accuracy of CRASH CAD in crash calculations, [7].

5.2 Crash analysis at the level of a single member.

Calculations at the cross - sectional level are the first

essential step in all CRASH CAD calculations. This section shows
how basic calculations, discussed in the previous section, are used
to predict crushing response of a single prismatic member.

5.2.1 Design of a section for axial crush.

One of the most sensitive parts of the design at the level of

a single prismatic section is concerned with progressive folding
during a head - on collision. Among all possible deformation
modes of a prismatic section the progressive folding absorbs most
of the impact energy. At the same time, this folding mode is the
one most difficult to obtain in a real - world design. Development
of progressive folding requires simultaneous completion of several
conditions. These are:

The cross - section topology must be properly designed, so

that, the local deformation of a section in each plastic lobe
can be accommodated without internal contacts and
penetrations. In addition, the deformation of each plastic
lobe must be compatible with the deformation of it's closest

Spot welds (rivets or laser weld - line) must not interfere
with the local plastic deformation of a section,

The section must be properly 'triggered' through the

introduction of correctly designed hoop dents which
guarantee the development of a proper 'natural' folding
mode and reduce the peak load to such a level that the
potentially unstable plastic deformations are induced only
in the region of triggering dents and finally

The boundary and loading conditions (stiffness of joints,

loading direction) are kept in the range that guarantees the
predominantly axial loading of the section.
The first three conditions, pertinent to the level of a single member,
must be met at the design stage of a given member while the last
condition must be checked at the level of full crash simulation of a
car. This difficult task requires an interaction of a component level
design tool with the full FE crash simulation model.

5.2.2 Design of the cross - sectional topology.

The design of a single member for axial crash is done in the

CRASH CAD design loop, [8]. The program leads the user through
several designs loops and suggests necessary corrections to the
cross - sectional geometry up to the point when the section can
collapse progressively without internal contacts and/or
penetrations. This stage of the design requires fine-tuning of
central angles, widths of side faces and appropriate geometry of
spot welding. An example of initial, bad design of a corrugated
panel and final, correct design is shown in Figure 20. It transpires
from Figure 20 that both cross-sections are quite similar and the
decision on the correctness of the design is impossible without a
detailed numerical simulation.

Figure 20 Initial (bad) and final (correct) topologies of a
corrugated panel optimized fro axial crash through a proper
selection of central angles and widths of side faces, [8].

5.2.3 Design of a triggering mechanism.

The final stage of cross-section design is concerned with

appropriate triggering mechanism. An example of triggering dents,
designed on the basis of SF modeling, are shown in Figure 21,
respectively, for hexagonal and octagonal (BMW) columns with

Figure 21 Paper models showing triggering dents in hexagonal and

octagonal (BMW) columns with flanges designed on the
basis of CRASH CAD calculations.

Introduction of triggering dents in members designed for
axial crash is necessary in order to promote a desired progressive
folding pattern and reduce the peak force below the level, which is
likely to induce a global, Euler - type buckling of a column. A
triggering of columns is especially important for complex
cross - sections that develop a large number of natural folding
modes. Usually only few of these modes are likely to converge to
the desired progressive folding pattern while other modes lead to a
premature bending of a column.

The importance of proper triggering is further clarified on an

example of a laboratory experiment on a simple cross - sectional
geometry, [9]. The first photograph in Figure 22 a shows a long
square prismatic column made of mild steel. Such a column has a
cross - sectional topology that guarantees proper folding without
internal contacts and does not contain spot welds that may destroy
progressive folding pattern. A properly triggered square column
collapses progressively up to the point when the last plastic fold is


Figure 22 Progressive collapse of a properly triggered long square
column, top, and global bending or irregular folding of
untriggered columns, bottom.

For example, the 54x54x1.4 mm square column at the top of Figure

22 a was squeezed over 700 mm and developed 32 symmetric
plastic lobes with no sign of any global bending (the completely
squeezed column in this figure was initially one-half of the size of
the longer column and developed 16 lobes). The total energy
absorbed was 25 kJ, which is sufficient to bring to the rest a 1000
kg car traveling with an initial velocity of 25 km/h. On the other
hand much shorter, untriggered, columns in Figure 22 b collapsed
in bending or deteriorated from progressive folding despite of a

carefully controlled loading and boundary conditions at both ends
of a column.

5.3 Structural components.

At the next level of a modeling process individual prismatic

components are assembled into more complex structural
assemblies. While the crushing characteristics as well as
interaction curves for each individual component are known from
the previous step of calculations the main difficulty at this stage of
design/calculation process is concerned with the construction of the
appropriate mechanical model for a given configuration of
structural components. The modeling complexity ar this stage is
similar to that encountered in the construction of a spring and mass
or multibody system models for given characteristics of
non - linear springs.

Simple models of structural assemblies are also developed by using

Superbeam Elements. Example of such a model is presented in the
next section.

5.3.1 The 2-D S-Frame model

The 'S' frame model is used to determine the crushing

response of planar deformation of S frame composed of prismatic
segments with an arbitrary cross - sectional topology. The entire 'S'
frame is discretized into only four Superbeam elements
(Superbeam is modeled by means of SF elements). Consequently,
the input procedure requires the specification of just overall
dimensions of a frame, as illustrated in Figure 23.

Figure 23 The input template of 'S' frame model. The whole frame is
discretized into four Superbeam Elements. The user is
asked to specify the dimensions shown in the figure.

The 'S' frame module calculates the peak force, energy absorption,
as well as the whole force - deflection characteristic for a given
frame. An example of calculations is shown in Figure 24 together
with corresponding experimental data taken from the SAE paper by
Y. Ohkami et al, [14].

Figure 24 Comparison of Crash Cad calculations with experimental

results reported in [6]. The energy absorption is calculated
for the crush of 200 [mm].

6 Conclusions

Present chapter addresses only a limited number of issues

concerned with macro element method. First of all it should be
recognized that despite all the unquestionable successes of
numerical simulation methods (e.g. FE) their limitations are
obvious. Designing of cars and/or predicting crushing response of
structures require intuitive judgment and qualitative
assessment - processes that cannot be precisely quantified.
Recognition of this fact lead to the development of the so-called
Qualitative Physics which involves methods such as commonsense
reasoning, expert-system-like techniques, diagrammatic reasoning
or hybrid methods, see e.g. [15]. The kinematic approach seems to
be a method that embraces these two recognized areas of structural
mechanics. The brief overview of the synthetic approach to the
simulation-based design process, discussed in the preceding
section, shows how both techniques could be combined into a
single integrated environment for crash simulation. In fact such
environments are currently under the development by a number of
car manufacturers. The related modeling methods and solution
techniques are not yet available in the open literature.

Secondly, a limited content of the present chapter does not

allow for the full exposition of the macro element method. The
results presented in sections 2 and 3 are limited to the basic
formulation of the method and basic examples of calculation
routines, [2], [6]. Problems such as bending and torsion crush
response [10],[11], interaction of cross-sectional forces [3], or
stability of collapse [6], [9], are not covered here. Interested
readers are referred to a number of papers and books listed at the
end of this chapter for more information regarding these subjects.

7 References.

[1] 1. J. M. Alexander, An approximate analysis of the collapse of

thin cylindrical shells under axial loading, Q. J. Mech. Appl.
Math., 13, 1, 10-15 (1960).
[2] W. Abramowicz and T. Wierzbicki, Axial crushing of multi-
corner sheet metal columns, J. App. Mech., 56, 1, 113-120 (1989).
[3] T. Wierzbicki and W. Abramowicz, Deep plastic collapse of
thin-walled structures, in Structural Failure (T. Wierzbicki and N.
Jones Ed.'s), John Wiley, New York (1989).
[4] W. Abramowicz, Crush Resistance of 'T' 'Y' and 'X' Sections,
Joint MIT-Industry Program on Tanker Safety, Massachusetts
Institute of Technology, Report 24 (1994).
[5] Abramowicz, W., Jones, N., Dynamic Progressive Buckling of
Circular and Square Tubes., Int. J. Impact Engng., 4, 4, 243 - 270,
[6] Wierzbicki, T., Abramowicz, W. The Manual of
Crashworthiness Engineering, Vol. I - IV, Center for
Transportation Studies, Massachusetts Institute of Technology,
1987 - 1989.
[7] McGregor, I., J., et al, Impact Performance of Aluminium
Structures, in Structural Crashworthiness and Failure (Editors N.
Jones & T. Wierzbicki), Elsevier 1993.
[8] Crash Cad User's Manual, Impact Design, Europe, 1998.
[9] Abramowicz, W., Jones, N., Transition from static and
dynamic progressive buckling to global bending of thin - walled
tubes, prepared for publication in the International Journal of
Impact Engineering.
[10] Wierzbicki, T., Recke L., Gholami, and Abramowicz, W.,
Stress Profiles in Thin-Walled Prismatic Columns Subjected to
Crush Loading. Part I. Compression., Computers & Structures 51,
6, 611-623, (1994).
[11] Wierzbicki, T., Recke L., Gholami, and Abramowicz, W.,
Stress Profiles in Thin-Walled Prismatic Columns Subjected to
Crush Loading. Part II. Bending., Computers & Structures 51, 6,
625-641, (1994).

[12] L. E. Malvern, Introduction to the Mechanics of a
Continuous Medium, Prentice-Hall, New Jersey (1969).
[13] R. Hill, Extremal paths of plastic work and deformation, J.
Mech. Phys. Solids, 34, 5, 511-523 (1986).
[14] Ohkami, Y., et al., Collapse of Thin - Walled Curved Beam
with Closed - Hat Section - Part. 1: Study on Collapse
Characteristic, SAE paper 900460, (1990).
[15] Kleiber, M., Kulpa, Z., Computer-assisted hybrid reasoning
and physical systems., Computer Assisted Mechanics and
Engineering Sciences, 2, 165-186, (1995).